Newsletters
Google unveiled Gemini 4 Carbon on Wednesday, its newest artificial intelligence (AI) model, saying it delivers major improvements in coding, autonomous agent tasks, and cybersecurity defense.
The release comes two weeks after the company launched its Gemini 4 Argon model.
In a blog post, Google called Carbon its most powerful frontier model, citing state-of-the-art cybersecurity and long-horizon capabilities and the ability to handle multi-step tasks over extended periods.
Google attributed the rapid release timeline to its researchers and AI models “working hand in hand to advance capabilities faster.”
The company claimed that Gemini 4 Carbon ranks first on the updated version 5.0 of the Artificial Intelligence Index, an independent ranking of AI models, with a score of 59.
Based on user feedback from its previous release, Google said it improved the model across full-stack web development, scientific tasks, and multimodal understanding, claiming an 89-point jump to fourth place on the WebDev Arena leaderboard, a ranking of AI models on web development tasks.
In cybersecurity benchmarks, the tech giant claimed the model holds a “firm lead,” scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation and 71% on CWE-bench for patching.
“On internal benchmarks, Carbon drives a strong uplift in generating secure code, and in live coding environments with our own teams it shows reduced misalignment and fabrication rates,” the company said.
Addressing safety and alignment, Google claimed that internal testing showed Carbon cuts unsupported claims by approximately half compared to Argon, stating methodological caveats directly rather than fabricating missing evidence.
The company also claimed the model retrieves relevant sources and surfaces discrepancies to users when facing ambiguity, while reducing task overreach by 26% and under-service by 29% during multi-step workflows.
Google said Gemini 4 Carbon will first become available to cybersecurity defenders this week before rolling out more broadly to developers, businesses and consumers.
The company plans to offer introductory pricing of $2 per one million input tokens (units of text an AI model processes) and $10 per one million output tokens, with cached input tokens discounted by 95%, before standard pricing rises to $4 per one million input tokens and $20 per one million output tokens.
Set as preferred source
Set as preferred source
1 min read
1 min read
